-
safety
2025-11-25 ~
↺ mudou de posição
A new alignment framing: the goal is an AI robustly caring about sentient life broadly, not just human control.
«An AI robustly aligned to care about sentient life... easier to build than one caring about human life alone.»
-
agi timeline
2025-11-25 ~
↺ mudou de posição
Gives a more cautious horizon for AI with human-like continual learning: roughly 5-20 years.
«I think like 5 to 20 [years].»
-
compute
2025-11-25
↺ mudou de posição
Argues the real bottleneck is now ideas, not compute; there are more companies than ideas.
«The bigger problem is ideas, not compute.»
-
compute
2025-11-25
↺ mudou de posição
In the Dwarkesh Patel interview: the age of scaling (2020-2025) is ending; an 'age of research' begins.
«it was the age of scaling... it's back to the age of research again, just with big computers.»
-
other
2025-10-01
In a legal deposition explains his OpenAI departure: he had a big new vision better suited to a new company.
«Ultimately, I had a big new vision. And it felt more suitable for a new company.»
-
compute
2024-12-13
↺ mudou de posição
At NeurIPS 2024 declares pre-training as we know it will end: peak data reached, data is the 'fossil fuel of AI'.
«Pre-training as we know it will unquestionably end... we've achieved peak data.»
-
compute
2024-09-04 ~
↺ mudou de posição
SSI raises $1B; Sutskever signals intent to scale differently from prior approaches.
«Everyone just says scaling hypothesis. Everyone neglects to ask, what are we scaling?»
-
safety
2024-06-19
↺ mudou de posição
SSI's founding statement: safety and capabilities advance in tandem, safety stays ahead, insulated from commercial pressure.
«We approach safety and capabilities in tandem... making sure our safety always remains ahead.»
-
safety
2024-06-19
↺ mudou de posição
Announces leaving OpenAI to found Safe Superintelligence: one focus, one goal, one product — safe superintelligence.
«We will pursue safe superintelligence in a straight shot, with one focus, one goal, and one product.»
-
safety
2023-07-17 ~
In a safety talk he favors a collaborative approach, hoping superintelligence itself will help solve the challenges it creates.
«I hope we will have the superintelligence itself help us solve the challenges it creates»
-
safety
2023-07-05
As OpenAI co-lead of Superalignment: superintelligence is the most impactful yet most dangerous technology, and could arrive this decade.
«Superintelligence will be the most impactful technology humanity has ever invented»
